Bayesian Network Hidden Markov Chain Model For Learning Analysis

A comprehensive study on data analysis of online learning was carried out of network school's practical demands for improving teaching management efficiency based on network school's learning management system(LMS)and WEB log data.The study was implemented theoretically and empirically.In the theoretical research,the hidden Markov model and procedure logistic model based on Bayes network were proposed to solve analysis modelling problems combining covariant in static state and hidden variables in dynamic behavioral sequence in the prediction of dropout.Meanwhile,given the parameter estimation process,variance estimation and model evaluation methods for model parameters were discussed.In the empirical research,feature extraction and analytical investigation of online study were firstly conducted;evaluative features about dropout and benchmark students were extracted by data mining,so as to confirm that the important problems influencing network operation and students' learning efficiency were the recognition and precaution of dropouts,and the indicator system for warning dropouts in online learning was established.Then,the construction method and optimization model for the model warning dropouts in online learning were researched deeply.Firstly,a static precaution model for online learning dropouts was constructed based on the traditional Logistic regression with multi-layered observational variables,as this can recognize dropouts in the early and medium stages as well as low efficiency in learning.However,the model's accuracy for the dropouts in early stage was not satisfactory;although it was improved after the behavioral state variable “online learning initiative” was brought in,it still was not able to support dynamic precaution.Finally,a modelling approach in theoretical research was adopted;the hidden Markov process regression based on Bayes network was introduced in;hierarchical regression with six sequence hidden variables “online learning initiative” was achieved,so that risk for dropout in the early stage was accurately and dynamically warned,proving the efficiency of theoretical research results with effects.The application value lies in: satisfying common demands for finding out dropouts in online educational institutions early for accurate recognition and precaution,which lays the foundation for continuous improvement of dynamic model for dropout precaution and also provides new theoretical model,computing method and application model for solving similar problems of predicting complex behaviors.Meanwhile,it offers a powerful theoretical support and method to promote big data application under the internet+ background.The innovation of this research is mainly embodied in the theory and application of hidden Markov chain process logistic regression model based on Bayes network.On the aspect of theoretical innovation: a 4-node trilateral three-layer Bayes network was proposed based on the application of predicting online learning behaviors and the composite likelihood function for solving model was given out;the middle-level nodes were combined with hidden Markov process and regression model,solving the problem of discrete dynamic state being regression hidden variable in the prediction of network behavior.Then,logistic model in hidden Markov process was specifically combined,which provided parameter estimation and model evaluation method.On the aspect of application innovation,Bayes network hidden Markov process model was brought into learning analysis field for the first time to predict dropout behaviors(the key and difficult problems in online education),preferably solving those complex,changing and unpredictable problems which were also hard for observe in online learning behavior sequence;meanwhile,a series of methods and models suitable for online educational institutions were developed.